Curious what kinds of failures people are seeing once agents move beyond demos and into real production workflows.
Some patterns we kept running into while building AgentPulse:
context getting dropped between agents
workflows degrading without throwing errors
debugging failures across multiple tools/agents
no clear visibility into where cost spikes happen
Would love to hear:
What s been the hardest issue to debug in your agent workflows so far?